bool BBox::intersect(const Ray& ray, float *tnear, float *tfar) const {
// you may already have those values hanging around somewhere
const __m128
plus_inf = loadps(ps_cst_plus_inf),
minus_inf = loadps(ps_cst_minus_inf);
// use whatever's apropriate to load.
const __m128
box_min = loadps(&min),
box_max = loadps(&max),
pos = loadps(&ray.o),
inv_dir = loadps(&ray.inv_d);
// use a div if inverted directions aren't available
const __m128 l1 = mulps(subps(box_min, pos), inv_dir);
const __m128 l2 = mulps(subps(box_max, pos), inv_dir);
// the order we use for those min/max is vital to filter out
// NaNs that happens when an inv_dir is +/- inf and
// (box_min - pos) is 0. inf * 0 = NaN
const __m128 filtered_l1a = minps(l1, plus_inf);
const __m128 filtered_l2a = minps(l2, plus_inf);
const __m128 filtered_l1b = maxps(l1, minus_inf);
const __m128 filtered_l2b = maxps(l2, minus_inf);
// now that we're back on our feet, test those slabs.
__m128 lmax = maxps(filtered_l1a, filtered_l2a);
__m128 lmin = minps(filtered_l1b, filtered_l2b);
// unfold back. try to hide the latency of the shufps & co.
const __m128 lmax0 = rotatelps(lmax);
const __m128 lmin0 = rotatelps(lmin);
lmax = minss(lmax, lmax0);
lmin = maxss(lmin, lmin0);
const __m128 lmax1 = muxhps(lmax,lmax);
const __m128 lmin1 = muxhps(lmin,lmin);
lmax = minss(lmax, lmax1);
lmin = maxss(lmin, lmin1);
const bool ret = _mm_comige_ss(lmax, _mm_setzero_ps()) & _mm_comige_ss(lmax,lmin);
storess(lmin, tnear);
storess(lmax, tfar);
return ret;
}
inline bool ray_box_intersect(const box_t & b, const ray_t & ray, rayseg_t & rs) {
/* you may already have those values hanging around somewhere */
const __m128
plus_inf = loadps(ps_cst_plus_inf), minus_inf = loadps(ps_cst_minus_inf);
/* use whatever's apropriate to load. */
const __m128
box_min = loadps(&b.min), box_max = loadps(&b.max), pos =
loadps(&ray.pos), inv_dir = loadps(&ray.inv_dir);
/* use a div if inverted directions aren't available */
const __m128 l1 = mulps(subps(box_min, pos), inv_dir);
const __m128 l2 = mulps(subps(box_max, pos), inv_dir);
/* the order we use for those min/max is vital to filter out */
/* NaNs that happens when an inv_dir is +/- inf and */
/* (box_min - pos) is 0. inf * 0 = NaN */
const __m128 filtered_l1a = minps(l1, plus_inf);
const __m128 filtered_l2a = minps(l2, plus_inf);
const __m128 filtered_l1b = maxps(l1, minus_inf);
const __m128 filtered_l2b = maxps(l2, minus_inf);
/* now that we're back on our feet, test those slabs. */
__m128 lmax = maxps(filtered_l1a, filtered_l2a);
__m128 lmin = minps(filtered_l1b, filtered_l2b);
/* unfold back. try to hide the latency of the shufps & co. */
const __m128 lmax0 = rotatelps(lmax);
const __m128 lmin0 = rotatelps(lmin);
lmax = minss(lmax, lmax0);
lmin = maxss(lmin, lmin0);
const __m128 lmax1 = muxhps(lmax, lmax);
const __m128 lmin1 = muxhps(lmin, lmin);
lmax = minss(lmax, lmax1);
lmin = maxss(lmin, lmin1);
const bool ret =
_mm_comige_ss(lmax, _mm_setzero_ps()) & _mm_comige_ss(lmax, lmin);
storess(lmin, &rs.t_near);
storess(lmax, &rs.t_far);
return ret;
}
bool Sphere::hit(const Ray &ray, IntersectionInfo &ii) const
{
// Make a vector to avoid Point -> Vector casting below
const Point3 localRayOrigin((worldToObject * ray.origin).get128());
const Vector3 localRayOriginAsVec(localRayOrigin);
const Vector3 localRayDir ((worldToObject * ray.direction()).get128());
#ifdef SSE4
const __m128 a = dotps(localRayDir.get128(), localRayDir.get128());
const __m128 b = mulps(set1ps(2), dotps(localRayDir.get128(), localRayOriginAsVec.get128()));
const __m128 rv = set1ps(r);
const __m128 c = subps(dotps(localRayOriginAsVec.get128(), localRayOriginAsVec.get128()), mulps(rv, rv));
const float ar = a.m128_f32[0];
const float br = b.m128_f32[0];
const float cr = c.m128_f32[0];
// Solve quadratic
const float d = (subps(mulps(b,b), mulps(set1ps(4), mulps(a, c)))).m128_f32[0];
#else
const float a = dot(localRayDir, localRayDir);
const float b = 2.f * dot(localRayDir, localRayOriginAsVec);
const float c = dot(localRayOriginAsVec, localRayOriginAsVec) - r*r;
const float ar = a;
const float br = b;
const float cr = c;
// Solve quadratic
const float d = b*b - 4.f * a*c;
#endif
if (d < 0)
return false;
const float sqrtD = sqrt(d);
float q;
if (br < 0)
q = -0.5f * (br - sqrtD);
else
q = -0.5f * (br + sqrtD);
float t0 = q / ar;
float t1 = cr / q;
if (t0 > t1)
std::swap(t0, t1);
if (t0 > ray.maxT || t1 < 0)
return false;
float hit = t0;
if (t0 < 0)
{
hit = t1;
if (hit > ray.maxT)
return false;
}
ray.maxT = hit;
// Now that we have a hit fill the intersection info structure
const Point3 localHitP(localRayOrigin + ray.maxT * localRayDir);
ii.P = Point3((objectToWorld * localHitP).get128());
ii.Ng = Vector3(localHitP) * invr;
ii.Ng = Vector3((worldToObjectN * ii.Ng).get128());
ii.N = ii.Ng;
ii.Cs = color;
ii.Os = opacity;
const float invPi = 1.f / 3.141592654f;
ii.s = ::asinf(ii.Ng.getX()) * invPi + 0.5f;
ii.t = ::asinf(ii.Ng.getY()) * invPi + 0.5f;
return true;
}
void GSSetupPrimCodeGenerator::Depth()
{
if(!m_en.z && !m_en.f)
{
return;
}
if(!m_env.sel.sprite)
{
// GSVector4 t = dscan.p;
movaps(xmm0, xmmword[edx + 16]);
if(m_en.f)
{
// GSVector4 df = p.wwww();
movaps(xmm1, xmm0);
shufps(xmm1, xmm1, _MM_SHUFFLE(3, 3, 3, 3));
// m_env.d4.f = GSVector4i(df * 4.0f).xxzzlh();
movaps(xmm2, xmm1);
mulps(xmm2, xmm3);
cvttps2dq(xmm2, xmm2);
pshuflw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
pshufhw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
movdqa(xmmword[&m_env.d4.f], xmm2);
for(int i = 0; i < 4; i++)
{
// m_env.d[i].f = GSVector4i(df * m_shift[i]).xxzzlh();
movaps(xmm2, xmm1);
mulps(xmm2, Xmm(4 + i));
cvttps2dq(xmm2, xmm2);
pshuflw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
pshufhw(xmm2, xmm2, _MM_SHUFFLE(2, 2, 0, 0));
movdqa(xmmword[&m_env.d[i].f], xmm2);
}
}
if(m_en.z)
{
// GSVector4 dz = p.zzzz();
shufps(xmm0, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
// m_env.d4.z = dz * 4.0f;
movaps(xmm1, xmm0);
mulps(xmm1, xmm3);
movdqa(xmmword[&m_env.d4.z], xmm1);
for(int i = 0; i < 4; i++)
{
// m_env.d[i].z = dz * m_shift[i];
movaps(xmm1, xmm0);
mulps(xmm1, Xmm(4 + i));
movdqa(xmmword[&m_env.d[i].z], xmm1);
}
}
}
else
{
// GSVector4 p = vertices[0].p;
movaps(xmm0, xmmword[ecx + 16]);
if(m_en.f)
{
// m_env.p.f = GSVector4i(p).zzzzh().zzzz();
movaps(xmm1, xmm0);
cvttps2dq(xmm1, xmm1);
pshufhw(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
pshufd(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
movdqa(xmmword[&m_env.p.f], xmm1);
}
if(m_en.z)
{
// GSVector4 z = p.zzzz();
shufps(xmm0, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
if(m_env.sel.zoverflow)
{
// m_env.p.z = (GSVector4i(z * 0.5f) << 1) | (GSVector4i(z) & GSVector4i::x00000001());
static const float half = 0.5f;
movss(xmm1, dword[&half]);
shufps(xmm1, xmm1, _MM_SHUFFLE(0, 0, 0, 0));
mulps(xmm1, xmm0);
cvttps2dq(xmm1, xmm1);
pslld(xmm1, 1);
cvttps2dq(xmm0, xmm0);
pcmpeqd(xmm2, xmm2);
psrld(xmm2, 31);
pand(xmm0, xmm2);
por(xmm0, xmm1);
}
else
{
// m_env.p.z = GSVector4i(z);
cvttps2dq(xmm0, xmm0);
}
movdqa(xmmword[&m_env.p.z], xmm0);
}
}
}
void GSSetupPrimCodeGenerator::Color()
{
if(!m_en.c)
{
return;
}
if(m_env.sel.iip)
{
// GSVector4 c = dscan.c;
movaps(xmm0, xmmword[edx]);
movaps(xmm1, xmm0);
// m_env.d4.c = GSVector4i(c * 4.0f).xzyw().ps32();
movaps(xmm2, xmm0);
mulps(xmm2, xmm3);
cvttps2dq(xmm2, xmm2);
pshufd(xmm2, xmm2, _MM_SHUFFLE(3, 1, 2, 0));
packssdw(xmm2, xmm2);
movdqa(xmmword[&m_env.d4.c], xmm2);
// xmm3 is not needed anymore
// GSVector4 dr = c.xxxx();
// GSVector4 db = c.zzzz();
shufps(xmm0, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
shufps(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
for(int i = 0; i < 4; i++)
{
// GSVector4i r = GSVector4i(dr * m_shift[i]).ps32();
movaps(xmm2, xmm0);
mulps(xmm2, Xmm(4 + i));
cvttps2dq(xmm2, xmm2);
packssdw(xmm2, xmm2);
// GSVector4i b = GSVector4i(db * m_shift[i]).ps32();
movaps(xmm3, xmm1);
mulps(xmm3, Xmm(4 + i));
cvttps2dq(xmm3, xmm3);
packssdw(xmm3, xmm3);
// m_env.d[i].rb = r.upl16(b);
punpcklwd(xmm2, xmm3);
movdqa(xmmword[&m_env.d[i].rb], xmm2);
}
// GSVector4 c = dscan.c;
movaps(xmm0, xmmword[edx]); // not enough regs, have to reload it
movaps(xmm1, xmm0);
// GSVector4 dg = c.yyyy();
// GSVector4 da = c.wwww();
shufps(xmm0, xmm0, _MM_SHUFFLE(1, 1, 1, 1));
shufps(xmm1, xmm1, _MM_SHUFFLE(3, 3, 3, 3));
for(int i = 0; i < 4; i++)
{
// GSVector4i g = GSVector4i(dg * m_shift[i]).ps32();
movaps(xmm2, xmm0);
mulps(xmm2, Xmm(4 + i));
cvttps2dq(xmm2, xmm2);
packssdw(xmm2, xmm2);
// GSVector4i a = GSVector4i(da * m_shift[i]).ps32();
movaps(xmm3, xmm1);
mulps(xmm3, Xmm(4 + i));
cvttps2dq(xmm3, xmm3);
packssdw(xmm3, xmm3);
// m_env.d[i].ga = g.upl16(a);
punpcklwd(xmm2, xmm3);
movdqa(xmmword[&m_env.d[i].ga], xmm2);
}
}
else
{
// GSVector4i c = GSVector4i(vertices[0].c);
movaps(xmm0, xmmword[ecx]);
cvttps2dq(xmm0, xmm0);
// c = c.upl16(c.zwxy());
movdqa(xmm1, xmm0);
pshufd(xmm1, xmm1, _MM_SHUFFLE(1, 0, 3, 2));
punpcklwd(xmm0, xmm1);
// if(!tme) c = c.srl16(7);
if(m_env.sel.tfx == TFX_NONE)
{
psrlw(xmm0, 7);
}
// m_env.c.rb = c.xxxx();
// m_env.c.ga = c.zzzz();
movdqa(xmm1, xmm0);
pshufd(xmm0, xmm0, _MM_SHUFFLE(0, 0, 0, 0));
pshufd(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
movdqa(xmmword[&m_env.c.rb], xmm0);
movdqa(xmmword[&m_env.c.ga], xmm1);
}
}
void GSSetupPrimCodeGenerator::Texture()
{
if(!m_en.t)
{
return;
}
// GSVector4 t = dscan.t;
movaps(xmm0, xmmword[edx + 32]);
movaps(xmm1, xmm0);
mulps(xmm1, xmm3);
if(m_env.sel.fst)
{
// m_env.d4.st = GSVector4i(t * 4.0f);
cvttps2dq(xmm1, xmm1);
movdqa(xmmword[&m_env.d4.st], xmm1);
}
else
{
// m_env.d4.stq = t * 4.0f;
movaps(xmmword[&m_env.d4.stq], xmm1);
}
for(int j = 0, k = m_env.sel.fst ? 2 : 3; j < k; j++)
{
// GSVector4 ds = t.xxxx();
// GSVector4 dt = t.yyyy();
// GSVector4 dq = t.zzzz();
movaps(xmm1, xmm0);
shufps(xmm1, xmm1, (uint8)_MM_SHUFFLE(j, j, j, j));
for(int i = 0; i < 4; i++)
{
// GSVector4 v = ds/dt * m_shift[i];
movaps(xmm2, xmm1);
mulps(xmm2, Xmm(4 + i));
if(m_env.sel.fst)
{
// m_env.d[i].si/ti = GSVector4i(v);
cvttps2dq(xmm2, xmm2);
switch(j)
{
case 0: movdqa(xmmword[&m_env.d[i].si], xmm2); break;
case 1: movdqa(xmmword[&m_env.d[i].ti], xmm2); break;
}
}
else
{
// m_env.d[i].s/t/q = v;
switch(j)
{
case 0: movaps(xmmword[&m_env.d[i].s], xmm2); break;
case 1: movaps(xmmword[&m_env.d[i].t], xmm2); break;
case 2: movaps(xmmword[&m_env.d[i].q], xmm2); break;
}
}
}
}
}
void GSDrawScanlineCodeGenerator::TestZ(const Xmm& temp1, const Xmm& temp2)
{
if(!m_sel.zb)
{
return;
}
// int za = fza_base.y + fza_offset->y;
mov(ebp, dword[esi + 4]);
add(ebp, dword[edi + 4]);
// GSVector4i zs = zi;
if(!m_sel.sprite)
{
if(m_sel.zoverflow)
{
// zs = (GSVector4i(z * 0.5f) << 1) | (GSVector4i(z) & GSVector4i::x00000001());
static float half = 0.5f;
movss(temp1, dword[&half]);
shufps(temp1, temp1, _MM_SHUFFLE(0, 0, 0, 0));
mulps(temp1, xmm0);
cvttps2dq(temp1, temp1);
pslld(temp1, 1);
cvttps2dq(xmm0, xmm0);
pcmpeqd(temp2, temp2);
psrld(temp2, 31);
pand(xmm0, temp2);
por(xmm0, temp1);
}
else
{
// zs = GSVector4i(z);
cvttps2dq(xmm0, xmm0);
}
if(m_sel.zwrite)
{
movdqa(xmmword[&m_env.temp.zs], xmm0);
}
}
if(m_sel.ztest)
{
ReadPixel(xmm1, ebp);
if(m_sel.zwrite && m_sel.zpsm < 2)
{
movdqa(xmmword[&m_env.temp.zd], xmm1);
}
// zd &= 0xffffffff >> m_sel.zpsm * 8;
if(m_sel.zpsm)
{
pslld(xmm1, m_sel.zpsm * 8);
psrld(xmm1, m_sel.zpsm * 8);
}
if(m_sel.zoverflow || m_sel.zpsm == 0)
{
// GSVector4i o = GSVector4i::x80000000();
pcmpeqd(xmm4, xmm4);
pslld(xmm4, 31);
// GSVector4i zso = zs - o;
psubd(xmm0, xmm4);
// GSVector4i zdo = zd - o;
psubd(xmm1, xmm4);
}
switch(m_sel.ztst)
{
case ZTST_GEQUAL:
// test |= zso < zdo; // ~(zso >= zdo)
pcmpgtd(xmm1, xmm0);
por(xmm7, xmm1);
break;
case ZTST_GREATER: // TODO: tidus hair and chocobo wings only appear fully when this is tested as ZTST_GEQUAL
// test |= zso <= zdo; // ~(zso > zdo)
pcmpgtd(xmm0, xmm1);
pcmpeqd(xmm4, xmm4);
pxor(xmm0, xmm4);
por(xmm7, xmm0);
break;
}
alltrue();
}
}
void GSDrawScanlineCodeGenerator::Step()
{
// steps -= 4;
sub(ecx, 4);
// fza_offset++;
add(edi, 8);
if(!m_sel.sprite)
{
// z += m_env.d4.z;
if(m_sel.zb)
{
movaps(xmm0, xmmword[&m_env.temp.z]);
addps(xmm0, xmmword[&m_env.d4.z]);
movaps(xmmword[&m_env.temp.z], xmm0);
}
// f = f.add16(m_env.d4.f);
if(m_sel.fwrite && m_sel.fge)
{
movdqa(xmm1, xmmword[&m_env.temp.f]);
paddw(xmm1, xmmword[&m_env.d4.f]);
movdqa(xmmword[&m_env.temp.f], xmm1);
}
}
else
{
if(m_sel.ztest)
{
movdqa(xmm0, xmmword[&m_env.p.z]);
}
}
if(m_sel.fb)
{
if(m_sel.tfx != TFX_NONE)
{
if(m_sel.fst)
{
// GSVector4i st = m_env.d4.st;
// si += st.xxxx();
// if(!sprite) ti += st.yyyy();
movdqa(xmm4, xmmword[&m_env.d4.st]);
pshufd(xmm2, xmm4, _MM_SHUFFLE(0, 0, 0, 0));
paddd(xmm2, xmmword[&m_env.temp.s]);
movdqa(xmmword[&m_env.temp.s], xmm2);
if(!m_sel.sprite)
{
pshufd(xmm3, xmm4, _MM_SHUFFLE(1, 1, 1, 1));
paddd(xmm3, xmmword[&m_env.temp.t]);
movdqa(xmmword[&m_env.temp.t], xmm3);
}
else
{
movdqa(xmm3, xmmword[&m_env.temp.t]);
}
}
else
{
// GSVector4 stq = m_env.d4.stq;
// s += stq.xxxx();
// t += stq.yyyy();
// q += stq.zzzz();
movaps(xmm2, xmmword[&m_env.d4.stq]);
movaps(xmm3, xmm2);
movaps(xmm4, xmm2);
shufps(xmm2, xmm2, _MM_SHUFFLE(0, 0, 0, 0));
shufps(xmm3, xmm3, _MM_SHUFFLE(1, 1, 1, 1));
shufps(xmm4, xmm4, _MM_SHUFFLE(2, 2, 2, 2));
addps(xmm2, xmmword[&m_env.temp.s]);
addps(xmm3, xmmword[&m_env.temp.t]);
addps(xmm4, xmmword[&m_env.temp.q]);
movaps(xmmword[&m_env.temp.s], xmm2);
movaps(xmmword[&m_env.temp.t], xmm3);
movaps(xmmword[&m_env.temp.q], xmm4);
rcpps(xmm4, xmm4);
mulps(xmm2, xmm4);
mulps(xmm3, xmm4);
}
}
if(!(m_sel.tfx == TFX_DECAL && m_sel.tcc))
{
if(m_sel.iip)
{
// GSVector4i c = m_env.d4.c;
// rb = rb.add16(c.xxxx());
// ga = ga.add16(c.yyyy());
movdqa(xmm7, xmmword[&m_env.d4.c]);
pshufd(xmm5, xmm7, _MM_SHUFFLE(0, 0, 0, 0));
pshufd(xmm6, xmm7, _MM_SHUFFLE(1, 1, 1, 1));
paddw(xmm5, xmmword[&m_env.temp.rb]);
paddw(xmm6, xmmword[&m_env.temp.ga]);
movdqa(xmmword[&m_env.temp.rb], xmm5);
movdqa(xmmword[&m_env.temp.ga], xmm6);
}
else
{
if(m_sel.tfx == TFX_NONE)
{
movdqa(xmm5, xmmword[&m_env.c.rb]);
movdqa(xmm6, xmmword[&m_env.c.ga]);
}
}
}
}
// test = m_test[7 + (steps & (steps >> 31))];
mov(edx, ecx);
sar(edx, 31);
and(edx, ecx);
shl(edx, 4);
movdqa(xmm7, xmmword[edx + (size_t)&m_test[7]]);
}
void GSDrawScanlineCodeGenerator::Init(int params)
{
const int _top = params + 4;
const int _v = params + 8;
// int skip = left & 3;
mov(ebx, edx);
and(edx, 3);
// left -= skip;
sub(ebx, edx);
// int steps = right - left - 4;
sub(ecx, ebx);
sub(ecx, 4);
// GSVector4i test = m_test[skip] | m_test[7 + (steps & (steps >> 31))];
shl(edx, 4);
movdqa(xmm7, xmmword[edx + (size_t)&m_test[0]]);
mov(eax, ecx);
sar(eax, 31);
and(eax, ecx);
shl(eax, 4);
por(xmm7, xmmword[eax + (size_t)&m_test[7]]);
// GSVector2i* fza_base = &m_env.fzbr[top];
mov(esi, dword[esp + _top]);
lea(esi, ptr[esi * 8]);
add(esi, dword[&m_env.fzbr]);
// GSVector2i* fza_offset = &m_env.fzbc[left >> 2];
lea(edi, ptr[ebx * 2]);
add(edi, dword[&m_env.fzbc]);
if(!m_sel.sprite && (m_sel.fwrite && m_sel.fge || m_sel.zb) || m_sel.fb && (m_sel.edge || m_sel.tfx != TFX_NONE || m_sel.iip))
{
// edx = &m_env.d[skip]
shl(edx, 4);
lea(edx, ptr[edx + (size_t)m_env.d]);
// ebx = &v
mov(ebx, dword[esp + _v]);
}
if(!m_sel.sprite)
{
if(m_sel.fwrite && m_sel.fge || m_sel.zb)
{
movaps(xmm0, xmmword[ebx + 16]); // v.p
if(m_sel.fwrite && m_sel.fge)
{
// f = GSVector4i(vp).zzzzh().zzzz().add16(m_env.d[skip].f);
cvttps2dq(xmm1, xmm0);
pshufhw(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
pshufd(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
paddw(xmm1, xmmword[edx + 16 * 6]);
movdqa(xmmword[&m_env.temp.f], xmm1);
}
if(m_sel.zb)
{
// z = vp.zzzz() + m_env.d[skip].z;
shufps(xmm0, xmm0, _MM_SHUFFLE(2, 2, 2, 2));
addps(xmm0, xmmword[edx]);
movaps(xmmword[&m_env.temp.z], xmm0);
}
}
}
else
{
if(m_sel.ztest)
{
movdqa(xmm0, xmmword[&m_env.p.z]);
}
}
if(m_sel.fb)
{
if(m_sel.edge || m_sel.tfx != TFX_NONE)
{
movaps(xmm4, xmmword[ebx + 32]); // v.t
}
if(m_sel.edge)
{
pshufhw(xmm3, xmm4, _MM_SHUFFLE(2, 2, 2, 2));
pshufd(xmm3, xmm3, _MM_SHUFFLE(3, 3, 3, 3));
psrlw(xmm3, 9);
movdqa(xmmword[&m_env.temp.cov], xmm3);
}
if(m_sel.tfx != TFX_NONE)
{
if(m_sel.fst)
{
// GSVector4i vti(vt);
cvttps2dq(xmm4, xmm4);
// si = vti.xxxx() + m_env.d[skip].si;
// ti = vti.yyyy(); if(!sprite) ti += m_env.d[skip].ti;
pshufd(xmm2, xmm4, _MM_SHUFFLE(0, 0, 0, 0));
pshufd(xmm3, xmm4, _MM_SHUFFLE(1, 1, 1, 1));
paddd(xmm2, xmmword[edx + 16 * 7]);
if(!m_sel.sprite)
{
paddd(xmm3, xmmword[edx + 16 * 8]);
}
else
{
if(m_sel.ltf)
{
movdqa(xmm4, xmm3);
pshuflw(xmm4, xmm4, _MM_SHUFFLE(2, 2, 0, 0));
pshufhw(xmm4, xmm4, _MM_SHUFFLE(2, 2, 0, 0));
psrlw(xmm4, 1);
movdqa(xmmword[&m_env.temp.vf], xmm4);
}
}
movdqa(xmmword[&m_env.temp.s], xmm2);
movdqa(xmmword[&m_env.temp.t], xmm3);
}
else
{
// s = vt.xxxx() + m_env.d[skip].s;
// t = vt.yyyy() + m_env.d[skip].t;
// q = vt.zzzz() + m_env.d[skip].q;
movaps(xmm2, xmm4);
movaps(xmm3, xmm4);
shufps(xmm2, xmm2, _MM_SHUFFLE(0, 0, 0, 0));
shufps(xmm3, xmm3, _MM_SHUFFLE(1, 1, 1, 1));
shufps(xmm4, xmm4, _MM_SHUFFLE(2, 2, 2, 2));
addps(xmm2, xmmword[edx + 16 * 1]);
addps(xmm3, xmmword[edx + 16 * 2]);
addps(xmm4, xmmword[edx + 16 * 3]);
movaps(xmmword[&m_env.temp.s], xmm2);
movaps(xmmword[&m_env.temp.t], xmm3);
movaps(xmmword[&m_env.temp.q], xmm4);
rcpps(xmm4, xmm4);
mulps(xmm2, xmm4);
mulps(xmm3, xmm4);
}
}
if(!(m_sel.tfx == TFX_DECAL && m_sel.tcc))
{
if(m_sel.iip)
{
// GSVector4i vc = GSVector4i(v.c);
cvttps2dq(xmm6, xmmword[ebx]); // v.c
// vc = vc.upl16(vc.zwxy());
pshufd(xmm5, xmm6, _MM_SHUFFLE(1, 0, 3, 2));
punpcklwd(xmm6, xmm5);
// rb = vc.xxxx().add16(m_env.d[skip].rb);
// ga = vc.zzzz().add16(m_env.d[skip].ga);
pshufd(xmm5, xmm6, _MM_SHUFFLE(0, 0, 0, 0));
pshufd(xmm6, xmm6, _MM_SHUFFLE(2, 2, 2, 2));
paddw(xmm5, xmmword[edx + 16 * 4]);
paddw(xmm6, xmmword[edx + 16 * 5]);
movdqa(xmmword[&m_env.temp.rb], xmm5);
movdqa(xmmword[&m_env.temp.ga], xmm6);
}
else
{
if(m_sel.tfx == TFX_NONE)
{
movdqa(xmm5, xmmword[&m_env.c.rb]);
movdqa(xmm6, xmmword[&m_env.c.ga]);
}
}
}
}
}
float TfirFilter::vec_inner_prod_sse(const float *eax, const float *edi, int ecx)
{
__m128 xmm3,xmm4,xmm0,xmm1,xmm5,xmm6;
xorps (xmm3, xmm3);
xorps (xmm4, xmm4);
ecx-=8;// sub $8, %%ecx
if (ecx<0) goto //jb
mul8_skip;
mul8_loop:
movups (xmm0,eax);
movups (xmm1,edi);
movups (xmm5,16/sizeof(float)+eax);
movups (xmm6,16/sizeof(float)+edi);
eax+=32/sizeof(float);
edi+=32/sizeof(float);
mulps (xmm1,xmm0);
mulps (xmm6,xmm5);
addps (xmm3,xmm1);
addps (xmm4,xmm6);
ecx-=8;
if (ecx>=0) //jae
goto mul8_loop;
mul8_skip:
addps (xmm3,xmm4);
ecx+=4;
if (ecx<0) //jl
goto mul4_skip;
movups (xmm0,eax);
movups (xmm1,edi);
eax+=16/sizeof(float);
edi+=16/sizeof(float);
mulps (xmm1, xmm0);
addps (xmm3, xmm1);
ecx-=4;
mul4_skip:
ecx+=4;
goto cond1;
mul1_loop:
movss (xmm0,eax);
movss (xmm1,edi);
eax+=4/sizeof(float);
edi+=4/sizeof(float);
mulss (xmm1,xmm0);
addss (xmm3,xmm1);
cond1:
ecx-=1;
if (ecx>=0) // jae
goto mul1_loop;
movhlps (xmm4,xmm3);
addps (xmm3,xmm4);
movaps (xmm4,xmm3);
//FIXME: which one?
xmm4=_mm_shuffle_ps(xmm4,xmm4,0x55);// shufps $0x55, xmm4, xmm4
// shufps $33, xmm4, xmm4
addss (xmm3, xmm4);
float sum;
movss (&sum , xmm3);
return sum;
}
